This is a ROS package for sensor (gaussian) noise filtering. This package consists of different filters (e.g. moving average, exponential moving average, butterworth) and signal visualization tool.
- Moving Average (MA)
- Exponential Moving Average (EMA)
- Butterworth Filter
- Signal Visualization
- Installation
ma_filter.py subscribes to a sensor topic e.g. sensor/dvl and publishes the filtered data to the output topic e.g. sensor/dvl_filtered
The current filtered value is taken as the average of n past values (including the current value), where n is the window size, the weightage of n past datas in current filtered data is same (average). Larger window size reduces more gaussian noise but results in more lag.
input_sensor_topic(default"/sensor/dvl")output_sensor_topic(default"/sensor/dvl_filtered")window_size(default5)
rosrun (optional param specification at the end)
rosrun sensor_noise_filter ma_filter.py _window_size:=8
roslaunch with param specification, add this node to your launch file
<node name="ma_filter" pkg="sensor_noise_filter" type="ma_filter.py" output="screen">
<param name="input_sensor_topic" value="/sensor/dvl" />
<param name="output_sensor_topic" value="/sensor/dvl_filtered" />
<param name="window_size" value="5" />
</node>
ema_filter.py subscribes to a sensor topic e.g. sensor/dvl and publishes the filtered data to the output topic e.g. sensor/dvl_filtered
The current filtered value is taken as alpha*current_value + (1 - alpha)*last_filtered_value. Recent data takes more weightage (depends on alpha). alpha=1 means no filtering, only current data is considered, high alpha give more weightage to current and recent data. Low alpha reduces more gaussian noise but results in more lag.
input_sensor_topic(default"/sensor/dvl")output_sensor_topic(default"/sensor/dvl_filtered")alpha(default0.5)
rosrun (optional param specification at the end)
rosrun sensor_noise_filter ema_filter.py _alpha:=0.7
roslaunch with param specification, add this node to your launch file
<node name="ema_filter" pkg="sensor_noise_filter" type="ema_filter.py" output="screen">
<param name="input_sensor_topic" value="/sensor/dvl" />
<param name="output_sensor_topic" value="/sensor/dvl_filtered" />
<param name="alpha" value="0.5" />
</node>
butterworth.py subscribes to a sensor topic e.g. sensor/dvl and publishes the filtered data to the output topic e.g. sensor/dvl_filtered
Butterworth filter is a low pass filter, eliminating high frequency gaussian noise. Low cutoff reduces more noise but results in higher delay. High order results in steeper roll-off (more restrictive to frequency beyond cutoff), but more delay and computational cost. More theoretical details can refer to https://www.youtube.com/watch?v=dmzikG1jZpU&list=PLdciPPorsHunK-nh9PyXcA8rTy4m-TuOV.
input_sensor_topic(default"/sensor/dvl")output_sensor_topic(default"/sensor/dvl_filtered")cutoff(default10): Cutoff frequencyfs(default33): Sampling rate of sensor in Hzorder(default4): Butterworth filter's order
rosrun (optional param specification at the end)
rosrun sensor_noise_filter butterworth.py _order:=8 _fs:=50
roslaunch with param specification, add this node to your launch file
<node name="butterworth" pkg="sensor_noise_filter" type="butterworth.py" output="screen">
<param name="input_sensor_topic" value="/sensor/dvl" />
<param name="output_sensor_topic" value="/sensor/dvl_filtered" />
<param name="cutoff" value="5" />
<param name="fs" value="33" />
<param name="order" value="4" />
</node>
💡NOTE: Run plot_frequency.py to determine an appropriate
cutoff, runrostopic hz /sensor/dvlto determine appropriatefs.
Subscribes to a sensor topic to plot its frequency domain signal. The plot is shown on the screen after the script stopped.
input_sensor_topic(default"/sensor/dvl")fs(default33)
rosrun (optional param specification at the end), press Ctrl+C after some time (for data collection) to end the script and show frequency plot.
rosrun sensor_noise_filter plot_frequency.py _input_sensor_topic:="/imu" _fs:=50
Subscribes to clean, noisy, & multiple filtered signal topic to evaluate and visualize filters' performance. It dynamically subscribes to filtered topics with common prefix e.g. sensor/dvl_filter (check available filtered topics every 0.3s). The performance metrics (snr, lag) is printed on console & the plot is saved in image_plots/ after the script stopped. The lag of filtered signal is calculated and compensated before SNR calculation using cross correlation. This is to eliminate the effect of lag on SNR. SNR calculated in this way may suffer from inconsistent sampling rate. Resolution of lag calculation depends on sampling rate.
clean_sensor_topic(default"/sensor/dvl_clean")noisy_sensor_topic(default"/sensor/dvl")prefix(default"/sensor/dvl_filtered"): Prefix for filtered topicsinterval(default0.3): Interval to check available filtered topicsfs(default33): sampling rate of all topicsplot_aligned_filtered_signal(defaultFalse): Plot the aligned filtered signal with compensated lag on the same graph (mainly for debug use)plots_dir(defaultimage_plots): The save directory of plots, relative to the package path e.g./path/to/sensor_noise_filter/image_plots
rosrun (optional param specification at the end), press Ctrl+C after some time (for data collection) to end the script, show metrics & save plots.
rosrun sensor_noise_filter plot_dvl.py _prefix:="/sensor/imu/filtered"
roslaunch: use plot.launch
Run plot_dvl.py & filter script(s) at the same time to evaluate, visualize and compare the performance of different filters in the same graph. Inspect plot.launch to include desired filters for comparison.
Run command below and press Ctrl+C after some time (for data collection) to show metrics, save plots & shutdown ROS.
roslaunch sensor_noise_filter plot.launch
💡NOTE: SNR & lag calculated by
plot_dvl.pyare not absolutely accurate due to inconsistent sensor sampling rate/missing datapoints. Runplot_dvl.pyorplot.launchbefore publishing sensor data to ensure same data length/minimal data losses to increase snr & lag accuracy.
cd /path/to/your/ROS_workspace/src
git clone https://github.com/bk1021/sensor_noise_filter.git
cd ..
rosdep install --from-paths src --ignore-src -r -y
catkin_make
source devel/setup.bash